spark-membrane · fail-closed console over pinned OES repositories (SYNTHETIC)

Read this first: results that do not flatter OES32

NEGATIVEOn the locked NASA SMAP/MSL protocol, OES32 did not meet its pre-stated success criterion: it beat only EWMA on SMAP, and maxabs and CUSUM scored a higher pooled F1 (not significantly).Source: sparkainlp-x/oes-resilience@1ee533cd4360a6b1415f923c4e52cfc166b8917a README, v0.5.0 section
NEGATIVEOn the oes-telemetry-bench SYNTHETIC held-out set, max-abs matched OES32's detections (6/7 each) with fewer false-alarm episodes per normal hour (11.54 vs 23.08); the rates come from about 5 minutes of synthetic normal time and are highly uncertain.Source: sparkainlp-x/oes-telemetry-bench@d49472b7e12d96ec25e62ff600a94a2b5ae49209 reports/synthetic-demo-report.json
NEGATIVEThe multi-quantum-oes preregistered SYNTHETIC stress evaluation found no demonstrated advantage: the block score did not beat a simple max-abs baseline.Source: sparkainlp-x/multi-quantum-oes@132e0a7b2ae05c742d5b0d91be221cc0d2def106 README, Stress evaluation
UNRUNoes32-hls FPGA synthesis is UNRUN; its latency figure is a design target, not a measurement.Source: sparkainlp-x/oes32-hls@56a4cc29f0bfed84583fe9c21879daefdb2a116c README
UNRUNThe qldpc_decoder_cpp HLS kernel has no belief-propagation updates yet; its HLS synthesis and hardware-in-the-loop runs are UNRUN.Source: sparkainlp-x/qldpc_decoder_cpp@908c65dd6eb30f21ffab3348704001e4155151cd README

There is no NASA-beat claim here. The dual-engine split (explorer + auditor) is an external inspiration; the pinned repositories do not already implement it.

The seed-42 run

python3 -m spark_membrane demo --seed 42 · protocol spark-membrane-demo-v2 (SHA-256 69efc39ab63b4b11…, locked hash matches). ACCEPT requires protocol_hash ∧ residual ∧ sidecar_A ∧ sidecar_C0 ∧ sidecar_C1 ∧ sidecar_Cfold ∧ weighted; max-abs, EWMA and CUSUM are advisory.

36ACCEPT frames

4LATCH frames

7frames where engines disagree

1REPAIRED_IN_SIM (simulation only)

3LATCH_HELD

Pick a frame

LATCHACCEPT engines disagree

t = 0 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 19 → pass · weighted score = 0.0077 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.523 ≥ 3?	quiet
CUSUM	0.371 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 1 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0098 vs tolerance 0.08 at index 28 → pass · weighted score = 0.0072 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.452 ≥ 3?	quiet
CUSUM	0.000 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 2 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 22 → pass · weighted score = 0.0076 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.722 ≥ 3?	quiet
CUSUM	0.101 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 3 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 28 → pass · weighted score = 0.0077 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.011 ≥ 3?	quiet
CUSUM	0.444 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 4 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 22 → pass · weighted score = 0.0074 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.233 ≥ 3?	quiet
CUSUM	0.000 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 5 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0097 vs tolerance 0.08 at index 22 → pass · weighted score = 0.0073 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.489 ≥ 3?	quiet
CUSUM	0.003 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 6 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 6 → pass · weighted score = 0.0078 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.679 ≥ 3?	quiet
CUSUM	1.284 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 7 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 30 → pass · weighted score = 0.0075 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.613 ≥ 3?	quiet
CUSUM	0.900 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 8 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0098 vs tolerance 0.08 at index 13 → pass · weighted score = 0.0078 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.662 ≥ 3?	quiet
CUSUM	0.685 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 9 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 2 → pass · weighted score = 0.0078 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.728 ≥ 3?	quiet
CUSUM	0.517 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 10 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0096 vs tolerance 0.08 at index 30 → pass · weighted score = 0.0078 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	1.715 ≥ 3?	quiet
CUSUM	1.904 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 11 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 16 → pass · weighted score = 0.0074 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	1.459 ≥ 3?	quiet
CUSUM	1.548 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 12 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 10 → pass · weighted score = 0.0072 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.874 ≥ 3?	quiet
CUSUM	0.560 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 13 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 11 → pass · weighted score = 0.0074 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	1.147 ≥ 3?	quiet
CUSUM	0.807 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 14 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0087 vs tolerance 0.08 at index 21 → pass · weighted score = 0.0065 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.009 ≥ 0.5?	quiet
EWMA	0.993 ≥ 3?	quiet
CUSUM	0.433 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 15 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 1 → pass · weighted score = 0.0073 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.326 ≥ 3?	quiet
CUSUM	1.368 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 16 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 8 → pass · weighted score = 0.0079 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.266 ≥ 3?	quiet
CUSUM	0.876 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 17 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0097 vs tolerance 0.08 at index 13 → pass · weighted score = 0.0074 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.854 ≥ 3?	quiet
CUSUM	1.278 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 18 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0098 vs tolerance 0.08 at index 22 → pass · weighted score = 0.0073 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	1.022 ≥ 3?	quiet
CUSUM	1.343 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 19 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0097 vs tolerance 0.08 at index 29 → pass · weighted score = 0.0071 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.396 ≥ 3?	quiet
CUSUM	1.524 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 20 · LATCH · `single_channel_spike` · engines disagree

residual: firedsidecar: firedweighted: quietmax-abs: quietEWMA: firedCUSUM: fired

Two formulas, never blended. Residual R = 0.1996 vs tolerance 0.08 at index 13 → LATCH · weighted score = 0.1045 vs 0.5 → quiet.

Failing gating checks
check (engine)	index	value vs threshold
`residual`oes32-residual (normative R)	13	0.1996 > 0.08
`sidecar_A`oes32_engine Profile A: coherence A	13	0.1996 > 0.08
`sidecar_C1`oes32_engine Profile A: ODD symmetry C1	13↔29	0.1986 > 0.08
`sidecar_Cfold`oes32_engine Profile A: FOLD8 continuity Cfold	13↔14	0.1931 > 0.08

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.200 ≥ 0.5?	quiet
EWMA	3.453 ≥ 3?	fired
CUSUM	5.784 ≥ 5?	fired

Explorer: REPAIRED_IN_SIM (evidence class SYNTHETIC). one-index delta -0.183669 at index 13 passes every gating check (|delta| <= cap 0.25); original frame kept. Proposals tried: 3.

Accepted proposal #3: channel 13, delta -0.183669; the original frame is kept.

Shield gate: EXPLORER WRITE → REFUSED (only DECODEUR or GARDIEN may write; EXPLORER proposals stay in simulation).

t = 21 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 12 → pass · weighted score = 0.0075 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.504 ≥ 3?	quiet
CUSUM	0.340 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 22 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 21 → pass · weighted score = 0.0076 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.581 ≥ 3?	quiet
CUSUM	0.137 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 23 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0098 vs tolerance 0.08 at index 18 → pass · weighted score = 0.0069 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	1.084 ≥ 3?	quiet
CUSUM	0.668 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 24 · LATCH · `spike_above_explorer_cap` · engines disagree

residual: firedsidecar: firedweighted: quietmax-abs: quietEWMA: firedCUSUM: fired

Two formulas, never blended. Residual R = 0.4010 vs tolerance 0.08 at index 5 → LATCH · weighted score = 0.2089 vs 0.5 → quiet.

Failing gating checks
check (engine)	index	value vs threshold
`residual`oes32-residual (normative R)	5	0.4010 > 0.08
`sidecar_A`oes32_engine Profile A: coherence A	5	0.4010 > 0.08
`sidecar_C1`oes32_engine Profile A: ODD symmetry C1	5↔21	0.3987 > 0.08
`sidecar_Cfold`oes32_engine Profile A: FOLD8 continuity Cfold	4↔5	0.4078 > 0.08

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.401 ≥ 0.5?	quiet
EWMA	8.855 ≥ 3?	fired
CUSUM	13.481 ≥ 5?	fired

Explorer: LATCH_HELD (evidence class SYNTHETIC). no single-index delta within cap 0.25 passed the audit in 16 seeded proposals (cap was binding). Proposals tried: 16.

t = 25 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0100 vs tolerance 0.08 at index 3 → pass · weighted score = 0.0071 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.440 ≥ 3?	quiet
CUSUM	0.234 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 26 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 23 → pass · weighted score = 0.0075 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	1.543 ≥ 3?	quiet
CUSUM	1.718 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 27 · LATCH · `two_channel_fault` · engines disagree

residual: firedsidecar: firedweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.1515 vs tolerance 0.08 at index 27 → LATCH · weighted score = 0.0841 vs 0.5 → quiet.

Failing gating checks
check (engine)	index	value vs threshold
`residual`oes32-residual (normative R)	27	0.1515 > 0.08
`sidecar_A`oes32_engine Profile A: coherence A	27	0.1515 > 0.08
`sidecar_C0`oes32_engine Profile A: EVEN symmetry C0	2↔18	0.1557 > 0.08
`sidecar_C1`oes32_engine Profile A: ODD symmetry C1	11↔27	0.1491 > 0.08
`sidecar_Cfold`oes32_engine Profile A: FOLD8 continuity Cfold	26↔27	0.1543 > 0.08

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.152 ≥ 0.5?	quiet
EWMA	0.702 ≥ 3?	quiet
CUSUM	0.387 ≥ 5?	quiet

Explorer: LATCH_HELD (evidence class SYNTHETIC). no single-index delta within cap 0.25 passed the audit in 16 seeded proposals. Proposals tried: 16.

t = 28 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 4 → pass · weighted score = 0.0069 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	1.643 ≥ 3?	quiet
CUSUM	1.634 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 29 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0097 vs tolerance 0.08 at index 28 → pass · weighted score = 0.0077 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	2.377 ≥ 3?	quiet
CUSUM	2.905 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 30 · ACCEPT · `slow_common_drift` · engines disagree

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: firedCUSUM: fired

Two formulas, never blended. Residual R = 0.0339 vs tolerance 0.08 at index 17 → pass · weighted score = 0.0293 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.034 ≥ 0.5?	quiet
EWMA	16.060 ≥ 3?	fired
CUSUM	26.002 ≥ 5?	fired

Explorer not invoked (frame accepted).

t = 31 · ACCEPT · `slow_common_drift` · engines disagree

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: firedCUSUM: fired

Two formulas, never blended. Residual R = 0.0346 vs tolerance 0.08 at index 11 → pass · weighted score = 0.0293 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.035 ≥ 0.5?	quiet
EWMA	13.863 ≥ 3?	fired
CUSUM	22.605 ≥ 5?	fired

Explorer not invoked (frame accepted).

t = 32 · ACCEPT · `slow_common_drift` · engines disagree

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: firedCUSUM: fired

Two formulas, never blended. Residual R = 0.0343 vs tolerance 0.08 at index 30 → pass · weighted score = 0.0288 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.034 ≥ 0.5?	quiet
EWMA	13.389 ≥ 3?	fired
CUSUM	21.814 ≥ 5?	fired

Explorer not invoked (frame accepted).

t = 33 · ACCEPT · `slow_common_drift` · engines disagree

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: firedCUSUM: fired

Two formulas, never blended. Residual R = 0.0349 vs tolerance 0.08 at index 13 → pass · weighted score = 0.0294 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.035 ≥ 0.5?	quiet
EWMA	13.851 ≥ 3?	fired
CUSUM	22.585 ≥ 5?	fired

Explorer not invoked (frame accepted).

t = 34 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0094 vs tolerance 0.08 at index 19 → pass · weighted score = 0.0072 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.009 ≥ 0.5?	quiet
EWMA	0.323 ≥ 3?	quiet
CUSUM	0.039 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 35 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0099 vs tolerance 0.08 at index 13 → pass · weighted score = 0.0076 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.069 ≥ 3?	quiet
CUSUM	0.000 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 36 · LATCH · `global_common_shift`

residual: firedsidecar: firedweighted: firedmax-abs: firedEWMA: firedCUSUM: fired

Two formulas, never blended. Residual R = 0.6072 vs tolerance 0.08 at index 31 → LATCH · weighted score = 0.6024 vs 0.5 → alarm.

Failing gating checks
check (engine)	index	value vs threshold
`residual`oes32-residual (normative R)	31	0.6072 > 0.08
`sidecar_A`oes32_engine Profile A: coherence A	31	0.6072 > 0.08
`sidecar_C1`oes32_engine Profile A: ODD symmetry C1	15↔31	1.2092 > 0.08
`weighted`oes-resilience weighted score	31	0.6024 ≥ 0.5

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.607 ≥ 0.5?	fired
EWMA	335.877 ≥ 3?	fired
CUSUM	559.202 ≥ 5?	fired

Explorer: LATCH_HELD (evidence class SYNTHETIC). no single-index delta within cap 0.25 passed the audit in 16 seeded proposals (cap was binding). Proposals tried: 16.

t = 37 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0096 vs tolerance 0.08 at index 1 → pass · weighted score = 0.0074 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.010 ≥ 0.5?	quiet
EWMA	0.075 ≥ 3?	quiet
CUSUM	0.000 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 38 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0089 vs tolerance 0.08 at index 24 → pass · weighted score = 0.0069 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.009 ≥ 0.5?	quiet
EWMA	0.160 ≥ 3?	quiet
CUSUM	0.000 ≥ 5?	quiet

Explorer not invoked (frame accepted).

t = 39 · ACCEPT · `none`

residual: quietsidecar: quietweighted: quietmax-abs: quietEWMA: quietCUSUM: quiet

Two formulas, never blended. Residual R = 0.0088 vs tolerance 0.08 at index 13 → pass · weighted score = 0.0063 vs 0.5 → quiet.

Every gating check passed.

Advisory engines (never change the verdict)
engine	score vs threshold	state
max-abs	0.009 ≥ 0.5?	quiet
EWMA	0.182 ≥ 3?	quiet
CUSUM	0.000 ≥ 5?	quiet

Explorer not invoked (frame accepted).

Engine families per frame: ● fired, · quiet; underlined columns are frames where the engines disagree. Only the gating families (residual, sidecar, weighted) decide the verdict.
t	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32	33	34	35	36	37	38	39
verdict	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	L	·	·	·	L	·	·	L	·	·	·	·	·	·	·	·	L	·	·	·
residual	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●	·	·	·	●	·	·	●	·	·	·	·	·	·	·	·	●	·	·	·
sidecar	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●	·	·	·	●	·	·	●	·	·	·	·	·	·	·	·	●	·	·	·
weighted	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●	·	·	·
max-abs	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●	·	·	·
EWMA	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●	·	·	·	●	·	·	·	·	·	●	●	●	●	·	·	●	·	·	·
CUSUM	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●	·	·	·	●	·	·	·	·	·	●	●	●	●	·	·	●	·	·	·

EWMA and CUSUM are standardized on a separate 64-frame event-free calibration stream (calibration-seed42.jsonl, never audited) and restart after each alarm (author choice; upstream defines no reset). At t=30, 31, 32, 33 the gate ACCEPTED while advisory engines fired: the gating checks do not see that slow drift.

Fail-closed check and the 512-channel bus

Tampered protocol

protocol tolerance edited from 0.08 to 0.5 after the lock: SHA-256 0d1b61891f580816… → LATCH on protocol_hash; 6 other gating checks not evaluated; explorer LATCH_HELD (explorer disabled: protocol hash mismatch (fail-closed)).

512 channels = 16 native-32 blocks, no resampling → LATCH

0ok

1ok

2ok

3ok

4ok

5ok

6ok

7ok

8ok

9ok

10ok

11LATCH

12ok

13ok

14ok

15ok

block 11: LATCH — residual@ch7 (global channel 359), sidecar_A@ch7 (global channel 359), sidecar_C1@ch7 (global channel 359), sidecar_Cfold@ch6 (global channel 358); explorer REPAIRED_IN_SIM

Architecture: five planes, one new algorithm

PLANE 1
Contract spine
oes32-residual · oes32_engine · oes32-membrane-shield
Normative residual R, Profile A sidecar (A, C0, C1, Cfold), capability gate.
re-implemented, tested against pins
PLANE 2
Frame bus
oes-telemetry-bench
Native 32-channel JSONL; 512 channels = 16 blocks of 32, no resampling.
re-implemented parser
PLANE 3
Audit side (T=0)
oes-resilience + baselines
Gate: hash ∧ residual ∧ sidecar ∧ weighted. Advisory: max-abs, EWMA, CUSUM.
conjunction logic only
PLANE 4
Explorer
new in this repository
Seeded one-index delta, capped by the protocol, re-audited. REPAIRED_IN_SIM or LATCH_HELD.
the only new algorithm
PLANE 5
Evidence membrane
measurement-trail · evidence-passport · quantum-claims-passport
Hash-chained trail, evidence passport, claims gate that refuses overclaims.
glue only

Frames flow 2 → 3; a LATCH from 3 goes to 4, whose proposal is re-audited by 3 and refused by the capability gate in 1; every step is chained by 5. The thresholds in plane 3 are listed with their sources in docs/PROTOCOL.md.

Evidence

Evidence passport (manifest.json, evidence-passport schema v1, result label synthetic example)
Run trail: 98 hash-chained records (measurement-trail format), head 2d2998ff451d73c1…
Full results JSON · input frames · calibration frames · console output
Claims gate: 9 ledger claims admitted; 26/26 probe overclaims refused; 6/6 honest denials admitted.

What this is, and is not

A classical software simulation on generated numbers. It is not a medical device, control software or field evidence. The audit is a conjunction of independent deterministic checks, not a vote, and the two OES-32 formulas (normative residual and weighted score) are never blended.

Pinned repositories (not copied)

Repository	Commit	Plane	DOI	Role here
oes32-residual	`b77b612`	contract spine	10.5281/zenodo.22985520	Normative OES-32 residual (ADR-001): R = max_i \|y_i - x_i\|, fail iff R > tolerance.
oes32_engine	`d66025f`	contract spine	10.5281/zenodo.22985523	Profile A sidecar: SAFE = A and C0 and C1 and Cfold (coherence, EVEN/ODD symmetry, FOLD8 continuity); LATCH = not SAFE.
oes32-membrane-shield	`95bb63a`	contract spine	10.5281/zenodo.22999276	Capability gate (Ed25519, v2; this commit adds the weak-key fix). spark_membrane/shield.py re-implements its capability format and role table, verifies signatures with the optional cryptography extra and ports its small-order/non-canonical key and R checks; checked against the upstream library in CI.
oes-telemetry-bench	`d49472b`	frame bus	10.5281/zenodo.23175492	Native 32-channel JSONL frame contract; max-abs / EWMA / CUSUM baseline definitions; synthetic held-out result (max-abs 6/7 at 11.54/h vs OES32 6/7 at 23.08/h).
oes-resilience	`1ee533c`	audit	10.5281/zenodo.23071166	Weighted block score 0.45max\|x\| + 0.35RMS + 0.20*mean\|x\| (OES32Detector); v0.5.0 SMAP/MSL result: success criterion not met.
measurement-trail	`7ab6ec8`	evidence membrane	10.5281/zenodo.23067465	Hash-chained JSONL record format used for the run trail; CI verifies the committed demo trail with the upstream verifier.
evidence-passport	`e275bb0`	evidence membrane	10.5281/zenodo.23165143	Run-manifest schema v1; CI validates docs/passport/manifest.json with the upstream validator.
quantum-claims-passport	`880ddc9`	evidence membrane	10.5281/zenodo.23167801	Claim-type classification with fail-closed fencing; the model for spark_membrane/claims.py.
multi-quantum-oes	`132e0a7`	linked context	10.5281/zenodo.23113851	Cited for its preregistered SYNTHETIC stress evaluation (no demonstrated advantage over max-abs). Its toy simulations never enter the audit.
oes32-hls	`56a4cc2`	linked context	10.5281/zenodo.22985525	Cited for status only: FPGA synthesis UNRUN. Its squared-difference, >= rule is a different Profile A sidecar and is not used here.
qldpc_decoder_cpp	`908c65d`	linked context	10.5281/zenodo.22985527	Cited for status only: BP kernel is a scaffold; HLS synthesis and hardware runs UNRUN.
signal-loom	`750a806`	linked context	10.5281/zenodo.23173453	Art skin only (performance-art prototype, MIT). Linked from the page; nothing from it enters the audit.

Existing pages (linked, not absorbed)

evidence-passport — Static evidence passports, including the SMAP/MSL run with its criterion-not-met label.
quantum-claims-passport — Claim-type audit that classifies public claims against their sources.
spark-oes512-demo — Browser prototype of the 16 x 32 weighted block score (SYNTHETIC).
oes512-residual — 16 x OES-32 weighted latch reference and browser avatar (SYNTHETIC).
signal-loom — Art skin only: a performance-art prototype. Nothing from it enters the audit.

How to cite

Concept DOI (all versions): 10.5281/zenodo.23175490; v0.2.1 version DOI: 10.5281/zenodo.23241695; v0.2.0 version DOI: 10.5281/zenodo.23186985; v0.1.0 version DOI: 10.5281/zenodo.23175491. See CITATION.cff.

Brisson, J.-F. (2026). spark-membrane: a fail-closed console over pinned OES repositories
(SYNTHETIC research prototype, version 0.2.0) [Computer software]. Spark AI NLP.
https://doi.org/10.5281/zenodo.23175490

Read this first: results that do not flatter OES32

The seed-42 run

Pick a frame

t = 0 · ACCEPT · none

t = 1 · ACCEPT · none

t = 2 · ACCEPT · none

t = 3 · ACCEPT · none

t = 4 · ACCEPT · none

t = 5 · ACCEPT · none

t = 6 · ACCEPT · none

t = 7 · ACCEPT · none

t = 8 · ACCEPT · none

t = 9 · ACCEPT · none

t = 10 · ACCEPT · none

t = 11 · ACCEPT · none

t = 12 · ACCEPT · none

t = 13 · ACCEPT · none

t = 14 · ACCEPT · none

t = 15 · ACCEPT · none

t = 16 · ACCEPT · none

t = 17 · ACCEPT · none

t = 18 · ACCEPT · none

t = 19 · ACCEPT · none

t = 20 · LATCH · single_channel_spike · engines disagree

t = 21 · ACCEPT · none

t = 22 · ACCEPT · none

t = 23 · ACCEPT · none

t = 24 · LATCH · spike_above_explorer_cap · engines disagree

t = 25 · ACCEPT · none

t = 26 · ACCEPT · none

t = 27 · LATCH · two_channel_fault · engines disagree

t = 28 · ACCEPT · none

t = 29 · ACCEPT · none

t = 30 · ACCEPT · slow_common_drift · engines disagree

t = 31 · ACCEPT · slow_common_drift · engines disagree

t = 32 · ACCEPT · slow_common_drift · engines disagree

t = 33 · ACCEPT · slow_common_drift · engines disagree

t = 34 · ACCEPT · none

t = 35 · ACCEPT · none

t = 36 · LATCH · global_common_shift

t = 37 · ACCEPT · none

t = 38 · ACCEPT · none

t = 39 · ACCEPT · none

Fail-closed check and the 512-channel bus

Tampered protocol

512 channels = 16 native-32 blocks, no resampling → LATCH

Architecture: five planes, one new algorithm

Contract spine

Frame bus

Audit side (T=0)

Explorer

Evidence membrane

Evidence

What this is, and is not

Pinned repositories (not copied)

Existing pages (linked, not absorbed)

How to cite

t = 0 · ACCEPT · `none`

t = 1 · ACCEPT · `none`

t = 2 · ACCEPT · `none`

t = 3 · ACCEPT · `none`

t = 4 · ACCEPT · `none`

t = 5 · ACCEPT · `none`

t = 6 · ACCEPT · `none`

t = 7 · ACCEPT · `none`

t = 8 · ACCEPT · `none`

t = 9 · ACCEPT · `none`

t = 10 · ACCEPT · `none`

t = 11 · ACCEPT · `none`

t = 12 · ACCEPT · `none`

t = 13 · ACCEPT · `none`

t = 14 · ACCEPT · `none`

t = 15 · ACCEPT · `none`

t = 16 · ACCEPT · `none`

t = 17 · ACCEPT · `none`

t = 18 · ACCEPT · `none`

t = 19 · ACCEPT · `none`

t = 20 · LATCH · `single_channel_spike` · engines disagree

t = 21 · ACCEPT · `none`

t = 22 · ACCEPT · `none`

t = 23 · ACCEPT · `none`

t = 24 · LATCH · `spike_above_explorer_cap` · engines disagree

t = 25 · ACCEPT · `none`

t = 26 · ACCEPT · `none`

t = 27 · LATCH · `two_channel_fault` · engines disagree

t = 28 · ACCEPT · `none`

t = 29 · ACCEPT · `none`

t = 30 · ACCEPT · `slow_common_drift` · engines disagree

t = 31 · ACCEPT · `slow_common_drift` · engines disagree

t = 32 · ACCEPT · `slow_common_drift` · engines disagree

t = 33 · ACCEPT · `slow_common_drift` · engines disagree

t = 34 · ACCEPT · `none`

t = 35 · ACCEPT · `none`

t = 36 · LATCH · `global_common_shift`

t = 37 · ACCEPT · `none`

t = 38 · ACCEPT · `none`

t = 39 · ACCEPT · `none`